How an AI photo calorie estimate works, and where it goes wrong

What the feature actually does

Point the camera at a plate, or pick a photo from your library, and Nyamly sends the image to Google Gemini, which identifies the dish, breaks it into its likely ingredients, and estimates a portion size for each one. From that, it produces calories, protein, carbs and fat for the meal. The same underlying model also powers a text-only version: type a dish description instead of a photo, and it returns the same kind of estimate — described in-app as less precise than a photo, and shown as an approximation.

Before any of this happens, Nyamly shows a one-time consent screen explaining that the photo (or description) is sent to Google Gemini for the estimate; what is kept, and for how long, is in the privacy policy. The feature is entirely optional: logging a meal by typing it in manually, or by scanning its barcode, needs no photo, no AI call and no consent at all.

An estimate you review, not a number that's final

The result is a starting point, not a finished diary entry. Nyamly shows the dish it detected, its guessed portion count, and the ingredients behind the number — every one of them editable. You can correct the portion size, remove an ingredient that isn't actually there, or adjust one that's wrong, and only then does anything get saved to your log. Nothing from a photo estimate reaches your diary without that review step.

Free and Pro limits

Where a photo estimate is least reliable

Nyamly's own in-app copy says it plainly: "Estimates, not measurements. Photo estimates are approximate and often run low, especially for large portions." A camera sees a plate from one angle; it doesn't see everything that went into the dish. A few situations are reliably harder than others:

• Hidden oil or butter. Cooking fat that's been absorbed into the food rather than sitting visibly on top adds real calories a photo simply can't see.
• Portion size. Without something in frame to judge scale against, a plate photographed close-up and one photographed from further back can look deceptively similar in size.
• Mixed dishes. A stew, casserole or stir-fry blends its ingredients together, which makes it harder to tell how much of each one is actually in the bowl.
• Sauces and dressings. A thin visible layer of sauce can carry a disproportionate share of a dish's calories, and how much was actually used is easy to underestimate from a photo.

None of this is a reason to distrust every estimate — it's the reason the review step exists. If you know a dish was cooked with extra oil, or the portion looks bigger or smaller than what's shown, that's exactly what to correct before saving.

Try it, then correct it

The fastest way to see how it handles your own meals is to try it. For how Nyamly rates the products you scan by barcode, see how Nyamly scores a product.